A Study on Automatic Shot Change Detection

1 I n t r o d u c t i o n The increase in low-cost mass storage media, improved compression techniques and higher bandwidth has meant that the use of video in multimedia applications has seen a rapid increase in recent years. The use of digital video is becoming more widespread in areas such as video conferencing, "video-on-demand", multimedia authoring systems, geographic information systems, monitoring systems, education, training and law enforcement, to name a few. For applications such as these large video databases are created and stored. The ability to browse the stored video data or to retrieve the content of interest is an essential functional requirement of any video archiving system. Recently, a considerable effort has been directed to finding ways to describe the information contained in video sequences. Methods for video annotation and for indexing into large video databases have been suggested to facilitate efficient search and query applications. The methods implemented are collectively described as content-based video retrieval. A number of systems have been proposed for the retrieval of multimedia data. In [1] these systems were broadly put under four categories: query by content, iconic query, SQL query and mixed queries. In query by content, the data is structured so that the queries are based on images, tabular form, similarity retrieval or by the component features of the video such as the shape, colour, or texture. With iconic query, normally after an initial query, a selection of "look alike" icons is presented to the user and the subsequent processing is defined by the user's selection of the appropriate icon(s). SQL queries follow the textual database management model whereby keywords are used as the basis for the queries usually assisted by Boolean operators (AND, OR) which specify the relationship between the keywords. Mixed queries use mixtures of the above methods, e.g. combining icons and keywords together.